This subsection explores heterogeneous effects of parental expectations on teenage girls by using an alternative definition of deprivation. In Section 7.1 the empirical models use deprivation percentiles derived from the Income Deprivation affecting Children Index (IDACI) based on information of 2001. This section uses the eligibility criteria followed by the social programme EiC-EAZ to find the deprived population. To do so, I construct a comparable group who did not receive the programme at all, but had similar characteristics to the targeted population prior to the implementation of the programme. The treatment group is composed by only those teenagers that attended schools that received in 1999 (primary school) the EiC-EAZ programme.
As Section 6 discusses, the programme was implemented in 1999, 2000, and 2001 across some deprived schools. The fact that this programme did not tackle every deprived school gives the opportunity for constructing a control or comparable group. By considering pre- treatment variables for constructing homogeneous groups, I am able to assess the effect of parental expectations among comparable groups facing similar peer and neighbourhood effects. It is worth noticing that this subsection does not use the information of the programme as an instrument, as the previous section does.
Propensity Score Matching (PSM) techniques were used for creating a comparable group called deprived or eligible group by the programme. The pre-treatment variables used for constructing the propensity score are: average size of teacher classes in Key Stage 1 (1997-2nd and 3rd year of primary school); evaluation of the 1996-1999 period of the standards achieved by the teenager’s school; percentage of pupils achieving level 4 or above in English and Maths (1998) and Science (1999); as well as parental schooling. Several PSM techniques were used to identify a comparable group to the population benefited by the programme, however, Table 16 only presents the results derived from the weights constructed by the method of “nearest neighbour without replacement and caliper”.57 Cochran and Rubin (1973) suggest this method for avoiding “bad” matches by imposing a tolerance in the maximum distance between the propensity score of treated and untreated individuals.
57The techniques used for identifying the common support were: nearest neighbour without replacement and
caliper, nearest neighbour with replacement and caliper, one to one matching, kernel-based matching, and the Mahalanobis distance. Kernel and nearest neighbour matching identified a common support with an absolute bias between 2 to 4.5 per cent. The Mahalanobis distance matching, using the same covariates as the rest of the techniques, was the method that identified a control group with the largest absolute bias, an average of 13 per cent of bias and a maximum of 24 per cent. The identification of the common support in all cases used the same seed, as well as a random ordering of the sample. This section of the paper uses the Stata matching package version 2010 (June) programmed by Barbara Sianesi, Institute for Fiscal Studies (IFS).
This table shows a significant and negative effect of parental expectations on the likelihood of becoming pregnant and having the child for deprived girls. The marginal effects reported by pregnancy models are similar to the ones reported by the first set of instruments, and greater in motherhood models. If high parental expectations are scant in deprived areas, we expect that their marginal contribution will be higher for deprived teenage girls than for non-deprived. Even though this happens for motherhood decisions, the comparison of these models with those using the complete sample must be taken with caution given that maximum likelihood models are extremely sensitive to different samples and the inclusion of different variables in both specifications.
7.4 Tests of Exogeneity
This subsection provides two types of tests of exogeneity when using maximum likelihood techniques. The first type uses Wald tests for analysing if the error correlations of the seemingly unrelated probits are significant. The second uses a control function approach also to test the exogeneity of parental expectations.
Tables 17 and 18 show the values of the Fisher’s transformation of the error correlation and their significance. If the error correlation is significant would reveal that even after considering instrumental variables and preferences for occupation, parental expectations are still endoge- nous. These tables show that the values of the correlations between ML and SML are very similar in magnitude and in significance for single instrumental probits and in significance for bivariate instrumental probits.
For teenage pregnancy models, there are significant correlations between the unobservable component of pregnancy and the reduced form of low expectations in the IV probits, even after using instruments (see column (2) of Table 17). A similar finding is reveal by both bivariate probits after accounting for the endogeneity of high parental expectations. However, when the three probits are considered simultaneously, column (5) shows no correlation between pregnancy and high or low parental expectations. The same conclusion is derived for motherhood models. As we expect, the only significant correlation is between the reduced form of low and high expectations given that they belong to the same categorical distribution.
The Wald Tests reported in Tables 19 and 20 present single tests that confirm the significance of the error correlations of the previous tables. In addition, they present joint tests for the trivariate simulated probits. Looking at column (5) of both tables, I conclude that the joint test of the error correlation equal to zero between pregnancy-high expectations and pregnancy- low expectations is not rejected. The same is observed for the motherhood specifications. These tests confirm that the estimates of parental expectations are not endogenous.58
To complement the tests of exogeneity, I provide the results of a method suggested by Bur- nett (1997) and discussed in Wooldridge (2010). This method consists on explicitly introducing
58
Wald Tests from the models using the second set of instruments are not shown in Appendix, however, single and joint tests reveal the same conclusions discussed in this subsection.
the residuals of the reduce form of parental expectations, shown in equation (7), as an addi- tional control of equation (6). As Wooldridge (2010) discusses, this method provides a valid test of the exogeneity of parental expectations by using probit specifications. Hence, high and low parental expectations are modelled as a probit and their generalized residuals are predicted; then, these are plugged into the pregnancy and motherhood specifications. Table 21 shows the results of this exercise, plugging both low and high expectation residuals into both models. The residuals are not significant and high parental expectations are still significant. This result does not reject the null hypothesis of exogeneity in parental expectations as concluded by testing the error correlations. It is worth highlighting that this method only provides a valid test of exogeneity, but does not consistently estimate the average treatment effects.
8
Robustness Checks
This final section discusses two exercises for assessing the robustness of my findings in pregnancy and motherhood models. The first exercise explores a different method to account for the selection bias caused by the attrition of the survey. The second focuses on pregnancy models to evaluate if the absence of questions about pregnancy experiences in Wave 4 and Wave 5 bias the estimates of parental expectations.
The first exercise uses the specification for predicting attrition from Wave 1 to Wave 6 used for constructing the inverse probability weights. To assess the effect of parental expectations in teenage pregnancy and motherhood, I include the selection probit model as an additional equation in the simulated trivariate probit. Thus, four equations are estimated by SML where the fourth is a selection probit.
Table 22 shows the marginal effects of born to a teenage mother, Key Stage 2 z-scores, and parental expectations using the EiC-EAZ programme as an instrumental variable. These specifications do not use inverse probability weights, instead, the selection probit accounts for the potential attrition bias through the error correlations in the SML. The findings of this exercise confirms the significance of parental expectations for both outcomes, and the magnitudes of these effects are similar to those discussed in the previous section. The standard errors are slightly bigger when I follow this approach.
Finally, as I have discussed in Section 5, pregnancy questions were collected only for Wave 6 and Wave 7 of the LSYPE. For recovering information about pregnancy in Wave 4 and Wave 5, I use information about motherhood. As a consequence, one concern is that parental expectations might be reflecting their effect on motherhood and not on pregnancy. Because the maximum likelihood results are not comparable when models from different samples are compared, I cannot assess this concern by reducing the sample for Wave 6 and Wave 7. However, I simple analysis about the influence of early waves on parental expectations estimates can be considered by interacting wave dummies with expectation variables.
to the previous models, this specification has as reference category Wave 7 instead of Wave 4. If parental expectation estimates in pregnancy models are mainly driven by the measure of motherhood from Wave 4 and Wave 5, interactions with early waves would be significant.59 This table shows that most of the interactions are not significant with exception of the interaction of Wave 6 with high parental expectations. By exploring these interactions, I can conclude that the marginal effects of pregnancy models, previously discussed, are not entirely explained by the measure of pregnancy in Wave 4 and Wave 5. Also, the number of observations about motherhood are relatively small in Wave 4 and Wave 5 in comparison to the rest of waves.
To sum up, after analysing SML with instrumental variables and evaluating the robustness of the main findings, I can conclude that parental expectations have a significant influence on teenage motherhood for the cohort analysed in this study. In addition, parental expectations significantly affect the teenager’s likelihood of becoming pregnant in deprived areas. Tests of exogeneity do not reject that parental expectations are exogenous after considering instrumental variables in the empirical specifications.
9
Conclusions
In this paper I analyse the effect of parental educational expectations about school choices on the likelihood of being pregnant and becoming a teenage mother. Based on simulated maximum likelihood methods using two sets of instrumental variables, my findings shed light on the extent and significance of parental expectations on teenage pregnancy and motherhood in England for a cohort of young people between 2004 to 2010.
After considering the potential endogeneity of parental expectations, I find that high parental expectations decreases the likelihood of teenage pregnancy and motherhood. The effect is about half as being born to a teenage mother. By using two sets of instrumental variables, one affect- ing deprived population and the other affecting a wider population, the main findings suggest that parental expectations marginal effects are mainly capturing a local effect in pregnancy models. For teenage motherhood specifications, the evidence suggests that the estimated ef- fect captures an average treatment effect. Pregnancy specifications show that high parental expectations decrease by 4 per cent the likelihood of being pregnant in comparison to teenage girls having parents reporting middle expectations (likely and fairly likely). Similarly, teenage motherhood models reveal that high parental expectations decrease by 2 percent the likelihood of becoming a teenage mother in comparison to the reference group.
Additionally, the results highlight that for teenage girls with poor academic performance, negative values of the standardised English z-score in Key Stage 2, the effect of parental ex- pectations is larger than for the rest of the teenage girls. This salient result might be reflecting the absence of additional sources of motivation and expectations at home and at school. As a result, this absence gives high parental expectations a marginal contribution on teenagers’
59
fertility decisions much larger than the observed for the rest of the population.
Although this study does not evaluate any of the employment or training programmes for young people, the theoretical model and the discussion of the mechanisms of parental ex- pectations suggest that these programmes may have indirect impacts on teenage pregnancy and motherhood through the expansion of alternatives for teenagers. The expansion of the teenager’s choice set, as well as the information about how to obtain a job or how to be en- rolled into an apprenticeship, may change teenagers’ fertility decisions for those who do not want to enroll in Higher Education.
Finally, the relevance of these findings has encouraged the extension of this work by con- sidering structural modelling techniques to better understand the formation of expectations on fertility choices. This framework will help on understanding how parents and teenagers form expectations and when these expectations matter most for teenagers’ fertility decisions. This extension will also help to forecast teenagers’ fertility decisions when the state of the world changes. A second extension of this work will be the analysis of the effect of job and train- ing programmes for the young people on the teenager’s fertility decisions by using panel data containing several cohorts.
References
Elizabeth O Ananat and Daniel M Hungerman. The power of the pill for the next generation: oral contraception’s effects on fertility, abortion, and maternal and child characteristics. Re- view of Economics and Statistics, 94(1):37–51, 2012.
Peter Arcidiacono, V Joseph Hotz, and Songman Kang. Modeling college major choices using elicited measures of expectations and counterfactuals. Journal of Econometrics, 166(1):3–16, 2012.
Adam Ashcraft and Kevin Lang. The consequences of teenage childbearing. Technical report, National Bureau of Economic Research, 2006.
Adam Ashcraft, Iván Fernández-Val, and Kevin Lang. The consequences of teenage child- bearing: Consistent estimates when abortion makes miscarriage non-random. The Economic Journal, 2013.
Orazio Attanasio and Katja Kaufmann. Educational choices, subjective expectations, and credit constraints. Technical report, National Bureau of Economic Research, 2009.
Wendy Baldwin and Virginia S Cain. The children of teenage parents. Family planning per- spectives, 12(1):34–43, 1980.
Hugo Benitez-Silva and Debra S Dwyer. The rationality of retirement expectations and the role of new information. Review of Economics and Statistics, 87(3):587–592, 2005.
Zoë Bradshaw and Pauline Slade. The effects of induced abortion on emotional experiences and relationships: a critical review of the literature. Clinical Psychology Review, 23(7):929–958, 2003.
Michael J Brien, Gregory E Loya, and John V Pepper. Teenage childbearing and cognitive development. Journal of Population Economics, 15(3):391–416, 2002.
Nancy J Burnett. Gender economics courses in liberal arts colleges. The Journal of Economic Education, 28(4):369–376, 1997.
John Bynner, Mary Londra, and Gill Jones. The impact of government policy on social exclusion among young people. Technical report, Office of the Deputy Prime Minister, Social Exclusion Unit, September 2004.
Lorenzo Cappellari and Stephen P Jenkins. Multivariate probit regression using simulated maximum likelihood. The Stata Journal, 3(3):278–294, 2003.
Willard Cates Jr, Kenneth F Schulz, and David A Grimes. The risks associated with teenage abortion. New England Journal of Medicine, 309(11):621–624, 1983.
Arnaud Chevalier and Tarja K Viitanen. The long-run labour market consequences of teenage motherhood in britain. Journal of Population Economics, 16(2):323–343, 2003.
William G Cochran and Donald B Rubin. Controlling bias in observational studies: A review. Sankyha, pages 417–446, 1973.
Áureo De Paula, Gil Shapira, and Petra E Todd. How beliefs about hiv status affect risky behaviors: Evidence from Malawi. Journal of Applied Econometrics, 2013.
Adeline Delavande. Pill, patch, or shot? subjective expectations and birth control choice. International Economic Review, 49(3):999–1042, 2008.
Adeline Delavande and Hans-Peter Kohler. HIV/AIDS-related expectations and risky sexual behaviour in Malawi, February 2014. University of Essex and University of Pennsylvania.
Adeline Delavande, Xavier Giné, and David McKenzie. Measuring subjective expectations in developing countries: A critical review and new evidence. Journal of Development Economics, 94(2):151–163, 2011.
Department for Children, Families and Schools. Teenage pregnancy strategy: Beyond 2010. Technical report, Department of Health, 2010.
Jeff Dominitz, Charles F Manski, and Jordan Heinz. Social security expectations and retirement savings decisions. Technical report, National Bureau of Economic Research, 2002.
Education Funding Agency. The 16-19 bursary fund: Your questions answered. Technical report, Education Funding Agency, 2013.
John Ermisch and David J Pevalin. Who has a child as a teenager? ISER Working Papers, 30, 2003.
John Ermisch and David J Pevalin. Early motherhood and later partnerships. Journal of Population Economics, 18(3):469–489, 2005.
William N Evans and Robert M Schwab. Finishing high school and starting college: Do catholic schools make a difference? The Quarterly Journal of Economics, 110(4):941–974, 1995.
David M Fergusson, L John Horwood, and Elizabeth M Ridder. Abortion in young women and subsequent mental health. Journal of Child Psychology and Psychiatry, 47(1):16–24, 2006.
Baruch Fischhoff, Aandrew M Parker, Wändi Bruine de Bruin, Julie Downs, Claire Palmgren, Robyn Dawes, and Charles F Manski. Teen expectations for significant life events. The Public Opinion Quarterly, 64(2):189–205, 2000.
Jason M Fletcher and Barbara L Wolfe. Education and labor market consequences of teenage childbearing evidence using the timing of pregnancy outcomes and community fixed effects. Journal of Human Resources, 44(2):303–325, 2009.
Marco Francesconi. Adult outcomes for children of teenage mothers. The Scandinavian Journal of Economics, 110(1):93–117, 2008.
Trevor Friedman and Dennis Gath. The psychiatric consequences of spontaneous abortion. The British Journal of Psychiatry, 155(6):810–813, 1989.
M Garel, B Blondel, N Lelong, S Bonenfant, and M Kaminski. Long-term consequences of miscarriage: the depressive disorders and the following pregnancy. Journal of reproductive and infant psychology, 12(4):233–240, 1994.
Arline T Geronimus and Sanders Korenman. The socioeconomic consequences of teen child- bearing reconsidered. The Quarterly Journal of Economics, 107(4):1187–1214, 1992.
Alissa Goodman, Greg Kaplan, and Ian Walker. Understanding the effects of early motherhood in britain: the effects on mothers. Discussion Paper 1131, IZA, 2004.
William H Greene. Econometric Analysis, 5th edition. Pearson Education India, 2003.
Jeffrey Grogger. Incarcelation among children of teen mothers. In Saul D Hoffman and Re- becca A Maynard, editors, Kids having Kids: Economic Costs and Social Consequences of Teen Pregnancy. The Urban Insitute Press, 2008.
Robert Haveman, Barbara Wolfe, and Elaine Peterson. The life chances of the children of teen mothers, 1968-88. In Saul D Hoffman and Rebecca A Maynard, editors, Kids having Kids: Economic Costs and Social Consequences of Teen Pregnancy. The Urban Insitute Press, 2008. Saul D Hoffman, E Michael Foster, and Frank F Furstenberg Jr. Reevaluating the costs of
teenage childbearing. Demography, pages 1–13, 1993.
V Joseph Hotz, Charles H Mullin, and Seth G Sanders. Bounding causal effects using data from a contaminated natural experiment: analysing the effects of teenage childbearing. The Review of Economic Studies, 64(4):575–603, 1997.
Daniel Kahneman. Thinking, fast and slow. Macmillan, 2011.
Thomas J Kane and Douglas Staiger. Teen motherhood and abortion access. The Quarterly Journal of Economics, 111(2):467–506, 1996.
Michael P Keane. A computationally practical simulation estimator for panel data. Economet- rica, pages 95–116, 1994.
Lesley Kendall, Lisa O’Donnell, Sarah Golden, Kate Ridley, Stephen Machin, Simon Rutt, Sandra McNally, Ian Schagen, Costas Meghir, Sheila Stoney, et al. Excellence in cities: The national evaluation of a policy to raise standards in urban schools 2000-2003. Technical Report Research Report No. 675A, Department for Education and Skills, 2005.
Daniel Klepinger, Shelly Lundberg, and Robert Plotnick. How does adolescent fertility affect the human capital and wages of young women? Journal of Human Resources, pages 421–448, 1999.
David I Levine and Gary Painter. The schooling costs of teenage out-of-wedlock childbearing: analysis with a within-school propensity-score-matching estimator. Review of Economics and Statistics, 85(4):884–900, 2003.
Arthur Lewbel, Yingying Dong, and Thomas Tao Yang. Comparing features of convenient estimators for binary choice models with endogenous regressors. Canadian Journal of Eco- nomics/Revue canadienne d’économique, 45(3):809–829, 2012.
Shelly Lundberg and Robert D Plotnick. Adolescent premarital childbearing: Do economic incentives matter? Journal of Labor Economics, pages 177–200, 1995.
Charles F Manski. Adolescent econometricians: How do youth infer the returns to schooling? In Studies of supply and demand in higher education, pages 43–60. University of Chicago